(DP 2001-08) Two Decades of Vector Autoregression (VAR) Modeling: A Survey

Abstract

A vector autogregression (VAR) is defined as a vector of endogenous variables regressed against its own lags. VARs therefore are considered part of a general class of simultaneous equations models. By construction, VAR analysis allows us to examine over time the dynamic impacts of innovations to variables on others. The following is a survey of the literature of vector autoregressions (VARs) in the last twenty years since it was first used for policy analysis for Christopher Sims (1980). Identification of a VAR model initially involved the imposition of a recursive structure on models. Since authors using VAR models provided insufficient justification for using a recursive structure, VAR modeling was criticized as atheoretical. In the last decade, however, a number of authors have attempted to remedy the problem by introducing new structural identification techniques. This has enhanced the ability of VARs to model the economy. VAR studies have been used primarily to identify the impacts of aggregate demand and supply shocks on aggregate output, as well as to identify the channels and impacts of monetary policy. The frontiers of current VAR research focuses on open economy extensions, as well as on improving lag selection and estimation.